Arrangement for receiving and transmitting PCI-X data according to selected data modes

1. Field of the Invention

The present invention relates to integrated network devices having Peripheral Component Interconnect (PCI) bridges.

2. Background Art

Peripheral Component Interconnect (PCI) interfaces have been used to provide high-speed connectivity between devices in a multi-device system, such as a processor based system such as a personal computer.

FIG. 1 is a diagram illustrating a conventional implementation of a PCI bus system architecture 100 . The system 100 includes a processor 102 coupled to a memory controller 104 via a local bus 106 . The processor 102 and the memory controller 104 are coupled to a PCI local bus 106 (labeled PCI Local Bus # 0 ) via a host bridge 108 .

The host bridge 108 provides a low latency path through which the processor 102 may directly access PCI devices 110 , for example a network interface card 110 a providing access to a local area network, a disc drive (SCSI) controller 110 b providing access to disk drives 114 , an audio card 110 c , a motion picture card 110 d , or a graphics card 110 e configured for driving a monitor 116 . The host bridge 108 also provides a high bandwidth path allowing PCI masters on the PCI bus 106 direct access to the system memory 118 via the memory controller 104 . A cache memory 120 is independent of the system memory 118 for use by the processor 102 .

The term “host bridge” refers to the bridge device 108 that provides access to the system memory 118 for the devices 110 connected to the PCI bus 106 . A PCI-to-PCI bridge 122 also may be used to connect a second PCI bus 124 to the PCI bus 106 , the second PCI bus 124 configured for connecting other I/O devices 126 .

Newer PCI bus protocols are being published, including PCI-X Mode 2, that provide enhanced PCI functionality. These newer PCI bus protocols include the PCI Local Bus Specification, Rev 2.3, the PCI-X Protocol Addendum to the PCI Local Bus Specification, Rev. 2.0a, and the PCI-to-PCI Bridge Architecture Specification, Rev 1.2.

The PCI-X Electrical and Mechanical Addendum to the PCI Local Bus Specification Revision 2.0a, Aug. 22, 2003 (hereinafter “PCI-X E/M Addendum”), specifically “Appendix C—PCI-X Mode 2 Device Design Guidelines and Examples” describes on pages 123-132 a proposal for implementing Double Data Rate (DDR) and Quad Data Rate (QDR) transmissions. In particular, the PCI data on the PCI-X bus is edge triggered (i.e., data is valid on the PCI-X bus when the PCI clock transitions from one state to another), such that the PCI data is latched on each rising and falling clock edge.

The implementation as suggested in the Appendix C suffers from the disadvantage that the disclosed technique for latching incoming data from the PCI-X bus creates numerous timing synchronization problems. According to the proposed implementation, multiple flip-flops inside the PCI device are arranged into an array for storing the incoming PCI data, where a selected flip-flop is used for latching a corresponding PCI data bit: decision logic outputs a signal used as the clock for latching the data into the selected flip-flop.

However, the proposed implementation as suggested in the Appendix C suffers from the disadvantage that the timing within the device is harder to predict, resulting in timing synchronization problems during implementation, for example due to implementation variations such as aligning strobe signals with the data, differences in board wire lengths, duty cycle variations, variations between drivers, and data pattern intersymbol interference.

There is a need for an arrangement that enables an integrated (i.e., single-chip) device, for example an integrated PCI bridge device, to transmit and receive PCI-X data according to double data rate (DDR) mode or quad data rate (QDR) mode, in an efficient manner that minimizes timing variations between clock signals and data signals.

These and other needs are attained by the present invention, where a PCI bridge device buffers received PCI bus strobe signals and distributes strobe signals for PCI receive data relative to the PCI data lines, enabling all PCI receive data to be latched using locally-generated PCI strobe signals generated based on the same PCI bus strobe signals. In addition, data line latch modules having primary and secondary flip-flops enable the PCI receive data to be held for an entire clock cycle, optimizing conversion between a PCI clock domain and a local clock domain of the PCI bridge device. A transmission circuit also can be configured to transmit data according to either double data rate (DDR) mode or quad data rate (QDR) mode in an efficient manner.

One aspect of the present invention provides an integrated PCI bridge device, the integrated PCI bridge device comprising first and second PCI interfaces. Each PCI interface includes a plurality of PCI bus latching modules configured for receiving respective prescribed groups of PCI data lines. Each PCI bus latching module has a plurality of data line latch modules, each data line latch module configured for latching from a corresponding PCI data line, according to the one data rate mode, first and second PCI data values based on respective locally-generated first and second PCI strobe signals. Each data line latch module further includes first and second buffers configured for generating the corresponding locally-generated first and second PCI strobe signals based on respective first and second PCI bus strobe signals received from the PCI bus. Each data line latch module is configured for holding the corresponding first and second PCI data values for at least a clock cycle of the respective locally-generated first and second PCI strobe signals. Each first and second PCI interface further includes means for outputting output data onto the corresponding prescribed group of PCI data lines according to the one data rate based on a data rate configuration signal.

Additional advantages and novel features of the invention will be set forth in part in the description which follows and in part will become apparent to those skilled in the art upon examination of the following or may be learned by practice of the invention. The advantages of the present invention may be realized and attained by means of instrumentalities and combinations particularly pointed in the appended claims.

Reference is made to the attached drawings, wherein elements having the same reference numeral designations represent like element elements throughout and wherein:

FIG. 1 is a block diagram of a conventional (PRIOR ART) implementation of a PCI based system.

FIG. 2 is a diagram of a PCI host bridge device according to an embodiment of the present invention.

FIG. 3 is a diagram illustrating a transmission circuit according to an embodiment of the present invention.

FIG. 4 is a timing diagram illustrating the transmission circuit of FIG. 3 operating according to double data rate (DDR) mode, according to an embodiment of the present invention.

FIG. 5 is a timing diagram illustrating the transmission circuit of FIG. 3 operating according to quad data rate (QDR) mode, according to an embodiment of the present invention.

FIG. 6 is a diagram illustrating a data line latch module configured for latching PCI data from a PCI bit line of the PCI bus, according to either DDR mode or QDR mode, according to an embodiment of the present invention.

FIG. 7 is a diagram illustrating enable signal generation circuits configured for generating the enable signals of FIG. 6 .

FIG. 8 is a diagram illustrating distribution of the strobe signals relative to the PCI data bit lines, according to an embodiment of the present invention.

The disclosed embodiment of the arrangement for using transmitting and receiving PCI-X data according to double data rate or quad data rate will be described in detail following an overview of the architecture of the HyperTransport™ tunnel/PCI-X bridge.

Architecture Overview

FIG. 2 is a diagram illustrating the transport tunnel-to-PCI host bridge device 10 , according to an embodiment of the present invention. The transport tunnel-to-PCI host bridge device 10 , preferably implemented as an integrated circuit on a single chip, is configured for operating as a HyperTransport™ (HT) tunnel device according to the HyperTransport™ IO Link Specification, Rev. 2.0, including errata up to specification Rev. 1.05c.

The transport tunnel-to-PCI host bridge device 10 includes two HT interfaces 12 a and 12 b labeled “Side 0 ” and “Side 1 ”, respectively. HT packets received by a given interface (e.g., 12 a ) from an HT bus 14 can be passed through the device 10 to the other interface (e.g., 12 b ) (i.e., tunneled), enabling multiple HT tunnel devices to be connected serially (e.g., a daisy-chain arrangement) to form an HT chain on the bus 14 for transfer of HT packets.

The following nomenclature may assist in describing connections in serially connected HT tunnel devices. The HT interface (e.g., 12 a ) that is connected to (or toward) a HyperTransport™ host device (e.g., a host memory controller) is referred to as the “upstream” interface, and the other HT interface (e.g., 12 b ) is referred to as the “downstream” interface: if the upstream interface (e.g., 12 a ) is directly connected to the HyperTransport™ host device via the corresponding connected HT tunnel 14 , then that tunnel connecting the upstream interface to the HyperTransport™ host device is referred to as the “upstream link”; if the upstream interface (e.g., 12 a ) is connected to a second HT device via an HT tunnel, where the second HT device provides a connection to (or toward) the HyperTransport™ host device, then that HT tunnel connecting the upstream interface and the second HT device is referred to as the “upstream tunnel”. Note that the downstream interface may or may not be connected to another HT device or chain of devices.

The transport tunnel-to-PCI host bridge device 10 also provides bridging operations between the HT bus 14 and PCI buses 16 a and/or 16 b . In particular, the transport tunnel-to-PCI host bridge device 10 contains two independent HT-to-PCI-X bridges 18 a and 18 b in accordance with the HyperTransport™ Link Specification, Rev. 2.0, the PCI Local Bus Specification, Rev 2.3, the PCI-X Protocol Addendum to the PCI Local Bus Specification, Rev. 2.0a, and the PCI-to-PCI Bridge Architecture Specification, Rev 1.2. The transport tunnel-to-PCI host bridge device 10 includes configurable registers ( 38 a , 38 b ) that identify the address ranges utilized by each of the PCI bus 16 a , the PCI bus 16 b , and the HT bus 14 .

Also note that the device 10 may identify a destination for a received HT packet based on a device identifier and bus identifier specified in the HT packet; alternately the device 10 may be configured to identify a destination bridge (e.g., 18 a , 18 b ) based on a corresponding unit identifier (e.g., 181 , 182 ) according to HyperTransport™ protocol. In addition, the transport tunnel-to-PCI host bridge device 10 includes configuration (CFG) registers 38 a , 38 b : the configuration register 38 a includes command and status registers (CSRs) within prescribed address spaces used by the transport tunnel-to-PCI host bridge device 10 for internal configuration, in accordance with the above-described PCI and HyperTransport™ specifications. Hence, the transport tunnel-to-PCI host bridge device 10 can store configuration information to an addressable CSR based on identifying an HT packet specifying a device identifier for the device 10 .

Hence, the transport tunnel-to-PCI host bridge device 10 can either tunnel HT packets received from the HT bus 14 by passing the packets from one interface (e.g., 12 a ) to the other interface (e.g., 12 b ), or the transport tunnel-to-PCI host bridge device 10 can pass the HT packets to one of the PCI buses 16 a or 16 b via the corresponding bridge 18 a or 18 b . Also note that the transport tunnel-to-PCI host bridge device 10 can process and respond to a received HT packet that is destined for the transport tunnel-to-PCI host bridge device 10 (e.g., for storage of configuration data).

The bridges 18 a and 18 b of the transport tunnel-to-PCI host bridge device 10 also forward onto the HT bus 14 any data having been received from the respective PCI buses 16 a or 16 b that specify a destination address within the prescribed address range specified within the CSRs for the HT bus 14 .

Each of the HT interfaces 12 a and 12 b includes a receive physical layer transceiver (RX PHY) 20 and a transmit physical layer transceiver (TX PHY) 22 . Each of the PHYs 20 and 22 include synchronization First-in-First-out (FIFO) registers for buffering packets according to a prescribed HT clock domain, and compensation circuitry to ensure electrical compliance with the HyperTransport™ specification. In particular, the RX PHY receives a packet and a clock associated with the packet according to the HT clock domain on the bus 14 ; the FIFO registers in the PHYs 20 and 22 are used to transition between the HT clock domain and a local clock (LCLK) domain 24 .

The device 10 includes a local clock (LCLK) domain 24 that relies on a clock that is independent of any variations in clocks driven by the HT bus 14 . The device 10 includes, within the local clock domain 24 , receive logic 26 and transmit logic 28 for each of the HT interfaces 12 and 12 b . The device also includes a phase locked loop (PLL) circuit 30 configured for generating the local clock (LCLK) and a phase-controlled clock (PCLK) 32 , and maintaining synchronization of those clocks as needed.

The receive logic 26 is configured for determining whether a received HT packet from the HT bus 14 should be forwarded to the corresponding transmit logic 28 for tunneling of the HT packet, whether the received HT packet should be forwarded to the bridge 18 for internal configuration of the device 10 , or forwarded to the bridge 18 for transfer of the received HT packet onto a determined one of the PCI buses 16 a or 16 b.

The device 10 also includes within the local clock domain a link interface cycle manager (GCM) 34 . The GCM 34 is configured for arbitrating the order of transactions that are to occur within the local clock domain 24 . For example, the GCM 34 may arbitrate between transactions that are to be output by the transmit logic (LTA) 28 . The reset logic (RST) 36 is configured for managing reset and device initialization operations, including implementing test and diagnostic modes.

The HT-to-PCI bridge 18 is configured for transferring HT packets, forwarded from the HT bus 14 by one of the receive logic blocks (e.g., LRA or LRB) 26 , to one of the PCI buses 16 a or 16 b . In particular, the receive logic block 26 having received the HT packet will specify to the bridge 18 the PCI bus (e.g., 16 a ) on which the HT packet is to be output. The HT-to-PCI bridge 18 also is configured for transferring packets from one of the PCI buses 16 a or 16 b to one of the transmit logic blocks (e.g., LTA or LTB) 28 for output onto the HT bus 14 . The bridge 18 (e.g., the bridge 18 a ) identifies the destination address of the HT packet as falling within the address range of either the PCI bus (e.g., 16 a ) or the configuration space of the configuration register (e.g., 38 a ), and forwards the packet accordingly.

The device 10 also includes, for each PCI bus (e.g., 16 a , 16 b ), a bridge 18 a and 18 b , the configuration registers 38 a and 38 b , a PCI interface module 40 , a PCI arbitration module 42 , and an interrupt controller 44 . The device 10 also includes a hot plug module 46 . The configuration registers 38 a and 38 b include registers, addressable via the HT bus 14 , for configuration of the respective components associated with bus operations for the respective PCI buses 16 a and 16 b . Each PCI interface module 40 is configured for transmitting and receiving PCI data bursts according to the above-identified PCI specifications. Each PCI arbitration module 42 is configured for managing PCI bus arbitrations for the corresponding attached PCI bus (e.g., 16 a or 16 b ), including identifying a bus master for PCI bus transactions. Each interrupt controller 44 is configured for servicing interrupts detected on the corresponding PCI bus (e.g., 16 a or 16 b ), as well as interrupts generated according to prescribed I/O Advanced Programmable Interrupt Controller (IOAPIC) specifications. Hence, various interrupt methods may be implemented in the interrupt controllers 44 .

The hot plug module 46 is configured for managing hot plug operations based on prescribed configuration registers in the registers 38 a and/or 38 b , where a selected PCI slot or device can be powered down while the device 10 , as well as other PCI devices on the PCI bus, remain in a powered state. For example, the hot plug module 46 may include logic that is compliant with the Standard Hot Plug Circuitry (SHPC) specified in the above-described PCI specifications; alternative implementations also may be included in the hot plug module 46 , for example logic configured for interfacing with commercially-available power controllers, including the TPS2340A or the TPS2342 power controllers commercially available from Texas Instruments, Inc.

Transmitting and Receiving PCI Data According to DDR or ODR Modes

FIG. 3 is a diagram illustrating a transmission circuit (i.e., an output cell) 50 configured for selectively outputting a corresponding PCI data bit onto the PCI-X bus (e.g., 16 a ) at either Double Data Rate (DDR) mode or Quad Data Rate (QDR) mode, according to an embodiment of the present invention. Each PCI interface 40 includes a plurality of cells 50 , each cell 50 configured for outputting data onto a corresponding bit line (or data pin) 55 . The transmission circuit 50 includes logic 52 configured for outputting clock signals CLKA, CLKB, and CLKD, and multiplexer selection signal LD. As illustrated in FIG. 4 , the logic 52 outputs the clock signals such that the clock signal CLKB leads the clock CLKD by a phase difference of 180°, and the clock signal CLKA leads the clock signal CLKB by a phase difference of 90° (clock signal CLKC is not used).

The transmission circuit 50 also includes multiplexers 54 , 56 , and 58 , and flip-flops 62 , 64 , 66 , 68 , and 70 .

The transmission circuit 50 is configured for receiving PCI data from the bridge device 10 as parallel inputs: the transmission circuit 50 receives the data inputs W, X in DDR mode, and the data inputs W, X, Y, Z in QDR mode.

The logic 52 is configured for causing the data inputs to be output onto a PCI data path 55 as a serial stream based on the PCI configuration (e.g., W, X in DDR, W, X, Y, Z in QDR). For example, in the Double Data Rate (DDR) configuration, only the data inputs W and X are supplied to the respective multiplexers 56 and 58 via the respective signal paths D 0 and D 4 . Hence, the logic 52 sets the mux selection input (LD) to “1”, such that the multiplexers 56 and 58 output the data inputs W and X from the signal paths D 0 and D 4 onto the signal paths D 2 and D 6 , respectively. In addition, in DDR configuration the logic 52 outputs the clock signals (CLKA, CLKB, and CLKD) at the PCI clock rate. The outputs D 2 and D 6 of the multiplexers 56 and 58 are synchronously latched by the flip-flops 64 and 68 , respectively, at the rising edge of the supplied clock signal (CLKD). The output D 7 of the flip-flop 68 is latched by the flip-flop 66 based on a second clock (CLKB).

Hence, the data value output by the flip-flop 66 onto signal path D 8 is delayed relative to the data value output by the flip-flop 64 onto signal path D 3 by half a clock cycle. Consequently, the selection input (CLKA) for the multiplexer 54 causes the data output (DOUT) to have a data rate that is twice the rate of the original input streams on input paths D 0 and D 4 .

As illustrated in FIG. 5 , when the logic 52 is set for performing Quad Data Rate (QDR) transmission, the logic 52 alternates the LD selection signal between “0” and “1” at the PCI clock rate, and the logic 52 outputs the clock signals (CLKA, CLKB, and CLKD) at twice the PCI clock rate. The logic 52 supplies the LD selection signal to the multiplexers 56 and 58 for selection between the data on data signal paths D 0 and D 1 by the multiplexer 56 , and the data on data signal paths D 4 and D 5 for the multiplexer 58 . Hence, the sequence of data supplied to the multiplexer 54 is controlled by the multiplexers 56 and 58 ; further, the sequence of input data (W, X, Y, Z) is further controlled by the flip-flops 64 , 66 , and 68 , where the data supplied on the input path D 8 is delayed relative to the input data on the D3 signal path, based on the flip-flop 66 being responsive to the CLKB signal which is 180° out of phase relative to the CLKB signal.

Hence, as illustrated below with respect to FIGS. 4 and 5 , the circuit 50 is configured for outputting data at double data rates, or quad data rates, as needed, while maintaining the same clock reference for all latching and outputting of data.

FIG. 4 is a diagram illustrating a timing diagram illustrating operation of the circuit 50 according to double data rate, according to an embodiment of the present invention. Each of the clock signals (CLKA, CLKB, and CLKD) are output by the logic circuit 52 according to the PCI bus clock frequency. In addition, the logic 52 sets the selection signal (LB) to a value of “1”, effectively disabling the inputs and D 1 and D 5 of the multiplexers 56 and 58 of FIG. 1 and setting the outputs D 2 and D 6 to equal the inputs D 0 and D 4 , respectively.

As illustrated in FIG. 4 , the data inputs D 0 and D 4 are supplied to the multiplexers 56 and 58 , respectively, each PCI clock cycle prior to the rising edge of the clock signal CLKD. The data inputs W 0 and X 0 are latched by the flip-flops 64 and 68 in response to the rising edge of the clock signal CLKD at event 200 . The output of the flip-flop 64 is supplied along the D3 signal path to the multiplexer 54 , whereas the output D 7 of the flip-flop 68 is latched by the flip-flop 66 in response to the rising edge of the clock signal (CLKB) at event 202 . The clock signal CLKA is used as the selection signal for the multiplexer 54 . Hence, the multiplexer 54 outputs at event 206 the data signal output from the flip-flop 64 (W 0 ), and outputs at event 208 the data output by the flip-flop 66 (X 0 ), based on the selection signal (clock signal CLKA).

Hence, the multiplexer 14 outputs the data supplied by the multiplexers 64 and 68 according to the sequence W 0 , X 0 , W 1 , X 1 , W 2 , X 2 , W 3 , X 3 , etc., at double the PCI clock rate based on the selection signal for the multiplexer 54 operating at the PCI clock rate. Hence, the data rate is doubled based on the two states of the clock being used to select two bits of data. In addition, the size of the data eye on bit line 55 is maximized based on the data on signal path D 3 being prepared while the data on signal path D 8 is driven, and vice versa. Hence, the selection signal (CLKA) transitions only while the input data is stable.

FIG. 5 is a timing diagram illustrating operation of the circuit 50 according to Quad Data Rate (QDR) mode, according to an embodiment of the present invention. As described above, the logic 52 is configured for outputting, in Quad Data Rate mode, the clock signals (CLKA, CLKB, and CLKD) at twice the PCI bus clock rate, and the mux selection signal (LD) at the same clock frequency as the PCI bus clock signal. Hence, the data inputs W, X, Y, and Z are supplied at the PCI bus clock rate which corresponds to the clock rate of the selection signal (LD). In the case of the D0 and D4 signal paths, QDR data transitions on signal paths D 0 and D 4 (e.g., from W 0 to W 1 on data path D 0 , and X 0 to X 1 on data path D 4 ) occur when the selection signal LD is low, and prior to the next rising edge of the CLKD signal, for example at event 300 . The supply of the PCI data Y, Z to the multiplexers 56 and 58 by the flip-flops 62 and 70 on signal paths D 1 and D 5 , respectively, are controlled by the rising clock edge of the CLKD signal (e.g., at event 302 ).

As illustrated in FIG. 5 , the data values W 0 and X 0 are output by the multiplexers 56 and 58 onto the respective signal paths D 2 and D 6 at event 302 in response to the LD selection signal being asserted at “1” value. Hence, the flip-flops 64 and 68 latch the respective data values W 0 and X 0 in response to the rising edge of the clock signal (CLKD) at event 302 , causing the flip-flops 64 and 68 to output the data values W 0 and X 0 onto the output paths D 3 and D 7 , respectively.

The data value W 0 is output by the multiplexer 54 onto the signal path DOUT 55 at event 304 in response to the selection signal (CLKA) switching from a “0” value to a “1” value. The multiplexers 56 and 58 also begin outputting the Y0 and Z0 values supplied on the D1 and D5 signal paths, respectively, in response to the LD selection input switching from a “1” value to a “0” value at event 304 .

The X0 value output by the flip-flop 68 onto signal path D 7 is latched at event 306 into the flip-flop 66 in response to the rising edge of the clock signal (CLKB), causing the flip-flop 66 to output the latched value X 0 onto the D8 signal path. Hence, the multiplexer 54 outputs the data value X 0 at event 300 in response to the selection input (CLKA) switching from a “1” value to a “0” value.

At event 308 , the flip-flops 64 and 68 latch the data values Y 0 and Z 0 output by the multiplexers 56 and 58 , respectively, enabling the multiplexer 54 to output the data value Y 0 at event 310 in response to the selection input CLKA switching from a zero (“0”) value to a one (“1”) value.

The data value Z 0 , having been output by the flip-flop 68 on the D7 signal path, is latched by the flip-flop 66 in response to the rising edge of the clock signal (CLKB) at event 312 , enabling the multiplexer 54 to output the data values Z 0 in response to the selection input (CLKA) switching from a one value (“1”) to a zero value (“0”) at event 314 .

At event 316 : the flip-flops 62 and 70 latch the data Y 1 and Z 1 , respectively, and the flip-flops 64 and 68 latch the values W 1 and X 1 output by the multiplexers 56 and 58 based on the LD selection value having a one (“1”) value. Hence, the latching of the data value W 1 by the flip-flop 64 enables the multiplexer 54 to output the W 1 value on the D3 signal path at event 318 in response to the selection input (CLKA) switching from a zero value (“0”) to a one value (“1”). In addition, the LD selection signal switches from a one value (“1”) to a zero value (“0”) at event 318 , causing the multiplexers 56 and 58 to output the data values supplied on inputs D 1 and D 5 , respectively.

As apparent from the foregoing, the sequence is repeated for the remaining data inputs, enabling the multiplexer 54 to output the data onto a single PCI data pin at four times the PCI clock rate.

According to the disclosed embodiment, receive logic is used to provide enable signals to selected flip-flops, enabling the selected flip-flops to prepare for receipt of the PCI data using the same clock signal.

FIGS. 6 , 7 , and 8 are diagrams illustrating the receive portion for a PCI interface 40 , according to an embodiment of the present invention. In particular, FIG. 8 illustrates the receive portion having input cells 800 a , 800 b , and 800 c . Each input cell (e.g., 800 a ) includes data line latch modules 500 for latching the respective data bits, and enable signal generation circuits 700 and 700 ′ for providing the enable signals for the associated data line latch modules 500 .

FIG. 6 illustrates a data line latch module 500 configured for latching a single bit 502 from the PCI bus. The data line latch module 500 is configured for converting the received serial data stream on the PCI bit line 502 into a parallel data word, and buffering between the PCI clock domain and the internal clock domain of the PCI bridge device.

Each of the primary flip-flops 504 a and 504 b , and the secondary flip-flops 508 of FIG. 6 are positive edge triggered flip-flops. The PCI data AD[x] supplied on the PCI input line 502 is latched by the primary flip-flop 504 a in response to a locally-generated PCI strobe signal “WF” on signal path 506 a . Data on the PCI input signal path 502 also is latched by the primary flip-flop 504 b in response to a locally-generated PCI strobe signal “WS” on signal path 506 b . As described below with respect to FIG. 8 , the locally-generated PCI strobe signals “WF” and “WS” on signal paths 506 a and 506 b are generated by strobe input cells 800 a and 800 b , respectively, that broadcast respective PCI bus strobe signals “FSTROBE” and “SSTROBE” received from the PCI-X bus.

The primary flip-flop 504 a outputs the latched value “dataF[x]” on data path 512 a to the secondary flip-flops 508 a , 508 b , 508 c , 508 d , 508 e , and 508 f ; the signal path 512 a also supplies the latched value “dataF[x]” to the multiplexers 510 a , 510 b , 510 c , 510 d , 510 e , and 510 f . Similarly, the primary flip-flop 504 b outputs the latched data dataS[x] onto data path 512 b to the secondary flip-flops 508 g , 508 h , 508 i , 508 j , 508 k , and 508 l ; the data path 512 b also supplies the latched value “dataS[x]” to the multiplexers 510 g , 510 h , 510 i , 510 j , 510 k , and 510 l.

Each of the first group of secondary flip-flops ( 508 a through 508 f ) has a corresponding enable signal EN F0-5 that is supplied by an enable signal generation circuit 700 , described below with respect to FIG. 7 . Each of the first group of multiplexers ( 510 a through 510 f ) also receives a corresponding one of the enable signals EN F0-5 as a selection input. Each of the second group of secondary flip-flops ( 508 g through 508 l ) has a corresponding enable signal EN S0-5 that is supplied by the enable signal generation circuit equivalent to the circuit 700 described below with respect to FIG. 7 . Each of the second group of multiplexers ( 510 g through 510 l ) also receives a corresponding one of the enable signals EN S0-5 as a selection input.

In addition, the first group of flip-flops ( 508 a through 508 f ) receive a strobe signal (datastrobeF[x]) on the strobe signal path 514 a , and the lower group of flip-flops ( 508 g through 508 l ) receive a strobe signal (datastrobeS[x]) via the signal path 514 b . Hence, each flip-flop 508 is configured for latching data in response to assertion of the corresponding enable signal on the enable signal path 520 a or 520 b , and based on its clock input 514 a or 514 b.

The strobe signals on strobe signal paths 506 a and 506 b , as well as the enable signals supplied on enable signal paths 520 a and 520 b , are generated by the strobe circuit 802 of FIG. 8 in a manner that ensures the incoming PCI data is latched, regardless of whether the data is supplied at a double data rate or quad data rate.

FIG. 7 is a diagram illustrating the enable signal generation circuit 700 for a corresponding portion of the data line latch module 500 . The enable signal generation circuit 700 generates the enable signals EN F0-5 for the first group of flip-flops ( 508 a through 508 f ) of FIG. 5 based on the strobe signal “strobeFb” on line 710 a . The circuit 700 includes a buffer 712 configured for buffering the strobe signal “strobeFb” and outputting the buffered strobe signal onto the signal path 714 .

A second enable signal generation circuit 700 ′, identical to the circuit 700 , is implemented for generation of the enable signals EN S0-5 for the second group of flip-flops ( 508 g through 508 l ) based on the strobe signal “strobeSb” on line 710 b . As described in the PCI-X E/M Addendum, the PCI bus strobe signal “FSTROBE” signal leads the PCI bus strobe signal “SSTROBE” by approximately a phase difference of 180°. Hence, the sequence of data AD[x] on the input signal path 502 is latched into the flip-flops according to the sequence 508 a , 508 g , 508 b , 508 h , 508 c , 508 i , 508 d , 508 j , 508 e , 508 k , 508 f , and 508 l , at which point the sequence is repeated.

The strobe enable signal generation circuit 700 is implemented as a shift register composed of flip-flops 702 a , 702 b , 702 c , 702 d , 702 e , and 702 f , and multiplexers 704 a , 704 b , 704 c , 704 d , 704 e , and 704 f . The flip-flops 702 a through 702 f are configured such that only one of the flip-flops (e.g., 702 a ) store a zero (“0”) value representing an enable signal, and the remaining flip-flops (e.g., 702 b through 702 f ) store a one (“1”) value representing a disable signal. Each of the flip-flops 702 a through 702 f latch the supplied data in response to the strobe signal “strobeFb” on signal path 714 if the corresponding enable signal has a zero (“0”) value.

Hence, the enable signal value (“0”) is passed among the flip-flops 702 a through 702 f , based on the strobe signal “strobeFb”, to enable the respective 508 a through 508 f flip-flops of FIG. 6 according to the appropriate sequence. The second strobe enable signal generation circuit 700 ′, identical to the circuit 700 , will be implemented for providing the enable signals EN S0-5 to the respective flip-flops 508 g through 508 l based on the strobe signal “strobeSb” supplied to the associated flip-flops of the second strobe enable signal generation circuit.

The multiplexers 704 a through 704 f are used to configure the circuit 700 for DDR or QDR mode. If the multiplexers 704 a through 704 f receive a selection signal “X” that is set to “1” in QDR mode (X=1), the multiplexers 704 a through 704 f cause each flip-flop 702 a through 702 d to receive an output from a corresponding preceding flip-flop (e.g., 702 a from 702 f , 702 b from 702 a , etc.). If the multiplexers 704 a through 704 f receive a selection signal “X” that is set to “0” in DDR mode (X=0), the multiplexers 704 a through 704 f cause the outputs of a first group of the flip-flops ( 702 a , 702 c , 702 e ) to bypass a second group of the flip-flops ( 702 b , 702 d , 702 f ).

Hence, in FIG. 6 all the flip-flops 508 are used in QDR mode, however in DDR mode only flip-flops 508 a , 508 c , 508 e , 508 g , 508 i , and 508 k are used.

Referring to FIG. 6 , the incoming PCI data AD[x] on the PCI bit line 502 are being supplied according to the PCI clock domain based on the incoming PCI bus strobe signals FSTROBE and SSTROBE, which are distributed throughout the PCI interfaces 40 as the strobe signals “WF” and “WS” on paths 506 a and 506 b , respectively. However, the PCI data AD[x] needs to be captured for latching and use in the local clock domain 513 of the PCI bridge device 10 .

Hence, conversion between the PCI clock domain and the local clock domain of the PCI bridge device 10 is accomplished by latching the PCI data AD[x] data into the flip-flops 510 , and holding the PCI data for latching by other flip-flops (not shown) within the device 10 according to the local clock domain.

The PCI data AD[x] input from the PCI signal path 502 is supplied to the flip-flops 504 a and 504 b via signal path 502 : the flip-flop 504 a latches the first data input on the signal path 502 based on the strobe signal “WF” on signal path 506 a , and the flip-flop 504 b latches the next data input on the signal path 502 based on the strobe signal “WS” on signal path 506 b . The data latched by the flip-flops 504 a and 504 b are supplied to the first group of flip-flops ( 508 a through 508 f ) and the second group of flip-flops ( 508 g through 508 l ), respectively. The first group of flip-flops ( 508 a through 508 f ) is latched by the data strobe “datastrobeF[x]” output by signal buffer 516 a onto strobe line 514 a , and second group of flip-flops ( 508 g through 508 l ) are latched by the data strobe “datastrobeS[x]” output by signal buffer 516 b onto strobe line 514 b.

At the end of the PCI-X data burst, the strobe signals on lines 506 a or 506 b for the last bit of data from the PCI burst on the data line 502 are used by the flip-flops 504 a or 504 b to latch the last bit of data. Since the PCI burst has ended, there will be no subsequent data strobe signal on lines 514 a or 514 b to transfer the latched data on lines 512 a or 512 b into the flip flops 508 . Hence, the multiplexers 510 are used to bypass the flip-flops 508 for the last PCI data bit latched by the flip-flops 504 a and 504 b.

As shown in FIG. 6 , at the time that a flip-flop (e.g., 508 a ) is enabled for latching data by its corresponding enable signal (e.g., EN F0 =0), the corresponding multiplexer (e.g., 510 a ) is configured for supplying the input of the corresponding flip-flop (e.g., 508 a ) to an output path 511 . As described above, the enable signal generation circuit 700 changes the enable values in response to a strobe signal “strobefb” on path 710 a ; hence, the multiplexer (e.g., 510 a ) outputs the input value until the next strobe signal that changes the corresponding enable signal value (e.g., EN F0 =0).

In response to the corresponding signal value being set to one (e.g., EN F0 =1), the multiplexer (e.g., 510 a ) outputs the data value (e.g., “dataF[x]”) output by the corresponding flip-flop (e.g., 508 a ). Hence, if a multiplexer (e.g., 510 a ) receives a zero value (e.g., EN F0 =0), the multiplexer (e.g., 510 a ) enters bypass mode. Only one multiplexer 510 is in bypass mode at any time. In other words, the multiplexer (e.g., 510 a ) associated with the corresponding flip-flop (e.g., 508 a ) is configured for selecting the output of the corresponding flip-flop only if there is valid latched data in the flip-flop. Hence, if the strobe signal “datastrobeF[x]” ceases before the flip-flop (e.g., 508 a ) has latched the input data “dataF[x]” on input path 512 a , the corresponding multiplexer (e.g., 510 a ) will still be able to propagate the data “dataF[x]” on the parallel data path 511 to the local clock domain 513 .

FIG. 8 is a diagram illustrating a PCI bus latching module 801 having input cells 800 a , 800 b , and 800 c configured for buffering and latching PCI strobe signals, the enable signal generation circuits 700 and 700 ′, and elasticity buffers 810 for latching PCI data bits for the local clock domain 513 , according to an embodiment of the present invention. The PCI bus latching module 801 of FIG. 8 is configured for latching sixteen bits (e.g., AD[ 0 ] through AD[ 15 ]) of a PCI bus; hence, the module 801 and the pair of strobe signals FSTROBE and SSTROBE are repeated four (4) times for a 64-bit PCI bus.

As illustrated in FIG. 8 , the input cells 800 a and 800 b are configured for buffering the strobe signals FSTROBE[ 0 ] and SSTROBE[ 0 ] and outputting the buffered signals WF and WS on signal paths 506 a and 506 b , respectively. The input cells 800 a and 800 b also output strobe signals “strobeFb” and “strobeSb” on paths 710 a and 710 b to the enable signal generation circuits 700 and 700 ′ based on the buffered signals WF and WS, respectively.

The enable signal generation circuits 700 and 700 ′ supply their respective enable outputs EN F and EN S via paths 520 a and 520 b to the sixteen (16) elasticity buffers 810 . Each elasticity buffer 810 in implementation includes the flip-flops 508 and the multiplexers 510 of FIG. 6 . Hence, each input cell 800 c in combination with a corresponding elasticity buffer 810 forms the data line latch module 500 of FIG. 6 .

Each input cell 800 a , 800 b , and 800 c includes a strobe buffer 802 configured for performing signal conditioning an incoming strobe signal (S_I). Each input cell 800 a , 800 b , and 800 c also includes the flip-flop 504 a , the flip-flop 504 b , the buffer 516 a , and the buffer 516 b of FIG. 6 . As shown in FIG. 8 , the input cells 800 a and 800 b are used to buffer the PCI bus strobe signals FSTROBE and SSTROBE from the PCI bus (e.g., 16 a ) via respective signal paths 804 a and 804 b , and generate the strobe signals “strobeFb” and “strobeSb”; hence the flip-flops 506 a and 504 b are unused in cells 800 a and 800 b.

Each input cell 800 c is used to latch a corresponding data bit AD[x] from the AD signal path 502 . The data bit is supplied via the signal paths 512 a , 512 b to the corresponding elasticity buffer 810 for latching based on the corresponding set of strobes on lines 514 a , 514 b.

Since an incoming PCI bus strobe signal (e.g., FSTROBE or SSTROBE) cannot be evenly distributed to match all the strobed data bits without introducing timing issues, the disclosed embodiment ensures that the PCI bus strobe signals FSTROBE and SSTROBE on respective signal paths 804 a and 804 b encounter minimal loading, as well as encountering the same propagation delays as the data on the AD[x] signal path 502 . Hence, the timing relationship between the input data and the corresponding strobe signals is preserved, without any errors introduced for example due to propagation delays relative to the data signal in the strobe signal.

Hence, buffer loading and strobe skew is minimized, since any localized delay encountered by a given input cell (e.g., 800 c ) is equal between the strobe cells and the input data. Although not shown, additional eight (8) additional input cells and elasticity buffers are used for capturing the eight (8) additional C/BE bits on the PCI bus 16 a , which are used as ECC checkbits during PCI-X mode 2 operation. These additional 8 input cells are distributed among the four (4) strobe groups.

The disclosed embodiment also reduces the number of clock domains down to three clock domains (namely the FSTROBE clock domain, the SSTROBE clock domain, and the local clock domains), as opposed to the PCI-X E/M Addendum which would have required 12 clock domains.

It will be readily apparent that the signal and register names are by way of illustration only, and that various modifications may be implemented consistent with the scope of the claims.

While this invention has been described with what is presently considered to be the most practical preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.